智能科学与技术学报 ›› 2023, Vol. 5 ›› Issue (2): 143-162.doi: 10.11959/j.issn.2096-6652.202315

• 综述与展望 • 上一篇    下一篇

行人轨迹预测方法关键问题研究:现状及展望

杜泉成1, 王晓2,3, 李灵犀4, 宁焕生1   

  1. 1 北京科技大学计算机与通信工程学院,北京 100083
    2 安徽大学人工智能学院,安徽 合肥 230601
    3 青岛智能产业技术研究院,山东 青岛 266109
    4 美国印第安纳大学-普渡大学印第安纳波利斯联合分校电子与计算机工程系,美国 印第安纳州 IN 46204
  • 修回日期:2023-03-24 出版日期:2023-06-15 发布日期:2023-06-10
  • 作者简介:杜泉成(1994- ),男,北京科技大学计算机与通信工程学院博士生,主要研究方向为轨迹预测和车辆规划决策
    王晓(1988- ),女,博士,安徽大学人工智能学院教授,青岛智能产业技术研究院院长,主要研究方向为社交网络分析、社交交通、网络运动组织和多智能体建模
    李灵犀(1977- ),男,博士,美国印第安纳大学−普渡大学印第安纳波利斯联合分校电子与计算机工程系教授,主要研究方向为复杂系统的建模、分析、控制与优化,互联与自动化车辆,智能交通系统,智能车辆,离散事件动态系统和人机交互
    宁焕生(1975- ),男,博士,北京科技大学计算机与通信工程学院教授,主要研究方向为网络空间及物联网研究
  • 基金资助:
    国家自然科学基金项目(U1811463);国家自然科学基金项目(62173329)

Key problems and progress of pedestrian trajectory prediction methods: the state of the art and prospects

Quancheng DU1, Xiao WANG2,3, Lingxi LI4, Huansheng NING1   

  1. 1 School of Computer and Communication Engineering, University of Science and Technology Beijing, Beijing 100083, China
    2 School of Artificial Intelligence, Anhui University, Hefei 230601, China
    3 Qingdao Academy of Intelligent Industries, Qingdao 266109, China
    4 Department of Electrical and Computer Engineering, Indiana University-Purdue University Indianapolis, Indianapolis IN 46204, USA
  • Revised:2023-03-24 Online:2023-06-15 Published:2023-06-10
  • Supported by:
    The National Natural Science Foundation of China(U1811463);The National Natural Science Foundation of China(62173329)

摘要:

行人轨迹预测旨在利用观察到的人类历史轨迹和周围环境信息来预测目标行人未来的位置信息,该研究具有重要的应用价值,可以降低自动驾驶车辆在社会交互下的碰撞风险。然而,传统的模型驱动的行人轨迹预测方法难以在复杂高动态的场景下对行人进行轨迹预测。相比之下,数据驱动的行人轨迹预测方法依靠大规模数据集平台,可以更好地捕捉和建模更复杂的行人交互关系,进而取得较精准的行人轨迹预测效果,成为自动驾驶、机器人导航和视频监控等领域的研究热点。为了宏观把握行人轨迹预测方法的研究现状及关键问题,以行人轨迹预测技术和方法分类为切入点,首先,详述行人轨迹预测已有方法的研究进展并归纳了目前存在的关键问题与挑战;其次,根据行人轨迹预测模型的建模差异,将现有方法分为模型驱动和数据驱动的行人轨迹预测方法,同时总结了不同方法的优缺点及适用场景;然后,对行人轨迹预测任务中使用的主流数据集进行了归纳总结,并对比了不同算法的性能指标;最后,针对行人轨迹预测的未来发展方向进行了展望。

关键词: 行人轨迹预测, 数据驱动, 社会交互, 自动驾驶

Abstract:

Pedestrian trajectory prediction aims to use observed human historical trajectories and surrounding environmental information to predict the future position of the target pedestrian, which has important application value in reducing collision risks for autonomous vehicles in social interactions.However, traditional model-driven pedestrian trajectory prediction methods are difficult to predict pedestrian trajectories in complex and highly dynamic scenes.In contrast, datadriven pedestrian trajectory prediction methods rely on large-scale datasets and can better capture and model more complex pedestrian interaction relationships, thereby achieving more accurate pedestrian trajectory prediction results, and have become a research hotspot in fields such as autonomous driving, robot navigation and video surveillance.In order to macroscopically grasp the research status and key issues of pedestrian trajectory prediction methods, We started with the classification of pedestrian trajectory prediction technology and methods.First, the research progress of existing pedestrian trajectory prediction methods were elaborated and the current key issues and challenges were summarized.Second, according to the modeling differences of pedestrian trajectory prediction models, existing methods were divided into model-driven and data-driven pedestrian trajectory prediction methods, and the advantages, disadvantages and applicable scenarios of different methods were summarized.Then, the mainstream datasets used in pedestrian trajectory prediction tasks were summarized and the performance indicators of different algoriths were compared.Finally, the future development direction of pedestrian trajectory prediction was prospected.

Key words: pedestrian trajectory prediction, data driven, social interaction, autonomous driving

中图分类号: 

No Suggested Reading articles found!